Artificial intelligence may be getting faster and sharper at forecasting everyday weather, but when the stakes are highest, it can still miss the mark.
A new study suggests that record-breaking weather events are slipping past some of today's most advanced AI weather models, including GraphCast, Pangu-Weather, and Fuxi.
With a database of extreme heat, cold, and wind events from 2018 to 2020, researchers used one of the world's leading physics-based weather prediction systems, called High Resolution Forecast, and leading AI models to produce weather forecasts and compared these predictions to real-world data.
The research, published in the journal Science Advances and detailed by Phys.org, found that leading AI forecasting systems performed worse than the physics-based model when it came to predicting the outcome of extreme weather.
According to the study, the AI models often outperformed HRES in routine daily forecasts and produced results much faster. But that advantage disappeared during record-setting events.
During major heat waves, for example, the AI systems repeatedly forecast temperatures below the levels that were actually observed. Even worse, as temperatures deviated more from records, the AI forecasts grew less accurate.
Get cost-effective air conditioning in less than an hour without expensive electrical work The Merino Mono is a heating and cooling system designed for the rooms traditional HVAC can't reach. The streamlined design eliminates clunky outdoor units, installs in under an hour, and plugs into a standard 120V outlet — no expensive electrical upgrades required. And while a traditional “mini-split” system can get pricey fast, the Merino Mono comes with a flat-rate price — with hardware and professional installation included. |
The scientists suggested the physics-based system's reliance on the laws of physics allowed it to predict record-breaking weather events more accurately than its AI counterparts. Even when the atmosphere behaves in unfamiliar ways, the laws of physics still apply.
AI models, on the other hand, rely on past data. When a weather event falls outside their training data, the models tend to pull predictions back toward more typical historical conditions instead of capturing the true severity.
Accurate forecasts inform the public and help protect lives, homes, power systems, and infrastructure as extremes driven by the changing climate become more common. If an extreme heat wave, deep freeze, or windstorm is underestimated, people may not be able to prepare in time.
Rather than ditching AI, the researchers pointed toward a blended path forward: Keep using AI for speed and efficiency, but pair it with physics-based systems for high-risk decisions.
TCD Picks » Quince Spotlight
💡These best-sellers from Quince deliver affordable, sustainable luxury for all
"Our findings underscore the current limitations of AI weather models in extrapolating beyond their training domain and in forecasting the potentially most impactful record-breaking weather events," they wrote. "... Further rigorous verification and model development is needed before these models can be solely relied upon for high-stakes applications such as early warning systems and disaster management."
Get TCD's free newsletters for easy tips, smart advice, and a chance to earn $5,000 toward home upgrades. To see more stories like this one, change your Google preferences here.
Cool Picks
